Musical expertise is related to altered functional connectivity during audiovisual integration

Evangelos Paraskevopoulosa,b, Anja Kraneburgb,c, Sibylle Cornelia Herholzd, Panagiotis D. Bamidisa, and Christo Pantevb,1

aSchool of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece; bInstitute for Biomagnetism and Biosignalanalysis, University of Münster, 48149 Münster, Germany; cResearch Group of , Leibniz Research Center for Working Enviroment and Human Factors, Dortmund, 44139 North Rhine-Westphalia, Germany; and dGerman Center for Neurodegenerative Diseases, 53175 Bonn, Germany

Edited by Josef P. Rauschecker, Georgetown University, Washington, DC, and accepted by the Editorial Board August 19, 2015 (received for review May 31, 2015) The present study investigated the cortical large-scale functional (8). A recent study by Paraskevopoulos et al. (9) used MEG to network underpinning audiovisual integration via magnetoencepha- reveal the cortical response related to identification of audiovi- lographic recordings. The reorganization of this network related to sual incongruences. These incongruences occurred on the basis long-term musical training was investigated by comparing musicians of an explicitly learned rule, binding otherwise unrelated uni- to nonmusicians. Connectivity was calculated on the basis of the sensory stimuli. The rule was comparable to musical reading: estimated mutual information of the sources’ activity, and the corre- “the higher the position of the circle, the higher the pitch of the sponding networks were statistically compared. Nonmusicians’ re- tone.” We recorded cortical responses to violations of this rule, sults indicated that the cortical network associated with audiovisual to incongruences between expected auditory and visual input. integration supports visuospatial processing and attentional shifting, This response was located in frontotemporal cortical areas. Plas- whereas a sparser network, related to spatial awareness supports the ticity effects related to long-term musical training were investigated identification of audiovisual incongruences. In contrast, musicians’ re- in this cross-sectional study by comparison of musicians and non- sults showed enhanced connectivity in regions related to the identi- musicians, whereas effects of short-term music reading training fication of auditory pattern violations. Hence, nonmusicians rely on were investigated in a follow-up longitudinal study that allowed

the processing of visual clues for the integration of audiovisual in- causal inference of the neuroplastic effects (10). formation, whereas musicians rely mostly on the corresponding au- A recent fMRI study by Lee and Noppeney (11) investigated ditory information. The large-scale cortical network underpinning how sensorimotor experience molds temporal binding of auditory multisensory integration is reorganized due to expertise in a cogni- and visual signals. The results of this study indicated that musicians tive domain that largely involves audiovisual integration, indicating exhibited increased audiovisual asynchrony responses and in- long-term training-related . creased connectivity selectively for music, and not for speech, in a superior temporal sulcus–premotor–cerebellar circuitry that was functional connectivity | MEG | multisensory integration | cortical defined and constrained by the corresponding fMRI activation plasticity | musical training results. Nevertheless, larger scale connectivity networks underpin- ning multisensory perception still remain undiscovered. ultisensory integration is of such importance for our un- The goal of the present study was to investigate the functional Mderstanding of the surrounding world that its cortical cor- connectivity of the cortical network underpinning audiovisual relates interact with most of the neocortical regions, including the ones traditionally considered as unisensory (1). The cortical areas Significance that are usually referred to as multisensory include the superior temporal sulcus, the intraparietal sulcus, and the frontal cortex (2). Multisensory integration engages distributed cortical areas and is Nevertheless, recent evidence suggests that multisensory percep- thought to emerge from their dynamic interplay. Nevertheless, tion engages more widespread areas than what the classic modular large-scale cortical networks underpinning audiovisual perception approaches have so far assumed (1). Moreover, audiovisual in- have remained undiscovered. The present study uses magneto- tegration emerges from a dynamic interplay of distributed regions encephalography and a methodological approach to perform operating in large-scale networks (3). whole- connectivity analysis and reveals, for the first The amount of shared information within these large-scale net- time to our knowledge, the cortical network related to multi- works of distributed neuronal groups conceptualizes the notion of sensory perception. The long-term training-related reorganiza- functional connectivity (4). Recent methodological advances allow tion of this network was investigated by comparing musicians the identification of networks that emerge from whole-brain analyses to nonmusicians. Results indicate that nonmusicians rely on of data such as functional magnetic resonance imaging processing visual clues for the integration of audiovisual in- (fMRI) and magnetoencephalography (MEG) (5) and provide fer- formation, whereas musicians use a denser cortical network that tile ground for studying the functional connectivity patterns of higher relies mostly on the corresponding auditory information. These cognitive processes. These kinds of networks are modeled as graphs data provide strong evidence that cortical connectivity is reor- composed of nodes, which represent the cortical areas contributing ganized due to expertise in a relevant cognitive domain, in- to the network, and by edges, which represent the connections be- dicating training-related neuroplasticity. tween the nodes. Each network has specific attributes that quantify connectivity organization (6). These graphs depict the functional Author contributions: E.P., A.K., S.C.H., P.D.B., and C.P. designed research; E.P. performed research; E.P. contributed new reagents/analytic tools; E.P. analyzed data; and E.P., A.K., of the respective cognitive processes. Recent evidence S.C.H., P.D.B., and C.P. wrote the paper. suggests that functional connectivity networks may be reorganized by The authors declare no conflict of interest. factors mediating neuroplasticity such as learning and development This article is a PNAS Direct Submission. J.P.R. is a guest editor invited by the Editorial (7), altering information processing pathways. Board. Music notation reading encapsulates auditory, visual, and 1To whom correspondence should be addressed. Email: [email protected]. motor information in a highly organized manner and therefore This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. provides a useful model for studying multisensory phenomena 1073/pnas.1510662112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1510662112 PNAS Early Edition | 1of6 Downloaded by guest on September 24, 2021 Fig. 1. Paradigm of an audiovisual congruent and incongruent trial. (A) A congruent trial. (B) An incongruent trial. The line “time” represents the duration of the presentation of the auditory and visual part of the stimulus. The last picture of each trial represents the intertrial stimulus in which subjects had to answer if the trial was congruent or incongruent.

integration, and particularly the identification of abstract audiovi- network of sources (P < 0.001; FDR corrected) compiled from 62 sual incongruences. Additionally, we aimed to address how this edges and 62 nodes, connecting mainly occipital (7 sources), pari- network is reorganized due to learning and whether expertise in a etal (12 sources), temporal (12 sources), and frontal (22 sources) cognitive domain that largely involves audiovisual integration, such areas (Fig. 2). In contrast to the musicians’ network, the non- as music reading, may relate to changes in whole-brain functional musicians’ network was lateralized mainly toward the left hemi- connectivity. To this aim we used MEG to measure the cortical sphere. The global efficiency of the network was found to be E = connectivity of a group of musicians and a group of musically naïve 0.00041, the density of the network was found to be D = 0.00016, controls to congruent and incongruent audiovisual stimuli (Fig. 1). whereas the node strength identified that the node with the greatest Source analysis was performed using a realistic head model to solve sum of strengths was located in the left middle frontal gyrus. the forward problem and standardized low resolution electromag- The between-groups comparison revealed that the congruent > netic tomography (sLORETA) (12) to solve the inverse one. The incongruent network of the musicians differed significantly (P < source time series were analyzed in terms of their underlying con- 0.001; FDR corrected) from the one of nonmusicians, showing nectivity, by measuring their mutual information (MI) and apply- greater connectivity in a network compiled from 68 edges and 49 ing a statistical model to extract the significant connections of the nodes, mainly connecting temporal (17 sources) and frontal network. This approach allowed us to identify the corresponding (18 sources) areas as well as parietal (7 sources) and occipital whole-head, node-to-node network following a graph theoretical (2 sources) ones (Fig. 2). The node with the greatest sum of approach. Our hypothesis for this analysis was that a widespread strengths was located in the right precentral gyrus. The sta- network connecting cortical areas previously identified to be im- tistical map of this analysis, along with the corresponding node portant for audiovisual integration would emerge. Regarding strengths, is presented in Fig. 2. training effects, we anticipated that musicians would show greater Cortical network underpinning audiovisual incongruency identification. The connectivity between distributed cortical areas and that the net- statistical analysis of the adjacency matrices of the group of musi- work characteristics would indicate enhanced processing efficacy cians with the planned contrast of incongruent > congruent compared with the nonmusicians’ network. revealed a complex network of sources (P < 0.05; FDR corrected) compiled from 85 edges and 71 nodes, mainly right lateralized, Results connecting occipital (11 sources), parietal (20 sources), temporal Behavioral Responses. The behavioral responses of all subjects were (3 sources), and frontal (29 sources) areas (Fig. 3). The global used to calculate the discriminability index d prime. One-sample efficiency of the network was found to be E = 0.00049, the t tests for each group revealed that both groups scored significantly density of the network was found to be D = 0.00022, whereas the higher than chance level {musicians: 2.897 [t (12) = 78.414; P < node strength identified that the node with the greatest sum of 0.001]; nonmusicians: 1.68 [t (12) = 6.527; P < 0.001]}. Moreover, strengths was located in the right postcentral gyrus. The analysis an independent-samples t test showed that musicians scored sig- of the adjacency matrices of the group of nonmusicians with the nificantly higher than the controls [t (24) = 4.678; P < 0.001], in- planned contrast of incongruent > congruent revealed a sparser dicating that musicians discriminated between the congruent and network of sources (P < 0.05; FDR corrected) compiled from incongruent trials with greater accuracy than the nonmusicians. 10 edges and 14 nodes, connecting mainly parietal (10 sources) The d prime results are presented in Fig. S1. and frontal (12 sources) areas but also occipital (2 sources), and temporal (4 sources) ones (Fig. 3). The global efficiency of the Graph Analysis Results. network was found to be E = 0.00026, the density of the network Cortical network underpinning audiovisual integration. The statistical was found to be D = 0.00003, whereas the node with the greatest analysis of the adjacency matrices of the group of musicians with sum of strengths was located in the left postcentral gyrus. the planned contrast of congruent > incongruent revealed a com- The between-groups comparison revealed that the incongruent > plex network of sources [P < 0.001; false discovery rate (FDR) congruent network of the musicians differed significantly (P < corrected] compiled from 73 edges and 60 nodes, mainly right lat- 0.05; FDR corrected) from the one of nonmusicians, showing eralized, connecting occipital (8 sources), parietal (10 sources), greater connectivity in a network compiled from 94 edges and temporal (19 sources), and frontal (16 sources) areas (Fig. 2). The 112 nodes, mainly connecting frontal (36 sources), temporal global efficiency of the network was found to be E = 0.00063, (25 sources), parietal (14 sources), and occipital (16 sources) the density of the network was found to be D = 0.00019, whereas areas (Fig. 3). The node with the greater sum of strengths was the node strength identified that the node with the greatest sum of located in the left inferior frontal gyrus. The statistical map of this strengths was located in the right precentral gyrus. The analysis of analysis, along with the corresponding node strengths are pre- the adjacency matrices of the group of nonmusicians with the sented in Fig. 3 and the network organization parameters are planned contrast of congruent > incongruent revealed a different presented in Table 1.

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1510662112 Paraskevopoulos et al. Downloaded by guest on September 24, 2021 Fig. 2. Cortical network underpinning audiovisual integration. (Upper) Statistical parametric maps of the significant networks for the congruent > in- congruent comparison. Networks presented are significant at P < 0.001, FDR corrected. The color scale indicates t values. (Lower) Node strength of the significant networks for each comparison. Strength is represented by node size.

Discussion regions. Sources located in superior and middle temporal gyrus The present study investigated the cortical large-scale functional contribute to a lesser extent, having direct connections with the network supporting audiovisual integration. The reorganization of middle occipital gyrus and the superior frontal one. The dorsal this network related to long-term musical training was investigated visual (15) pathway is thought to be responsible for the processing by comparing musicians to nonmusicians. Magnetoencephalo- of spatial information, receiving much of its input from magno-

graphic recordings were used to measure cortical responses. Large- cellular cells (16). The structural substrates of the dorsal auditory NEUROSCIENCE scale connectivity of the cortical activity was calculated via a sta- pathway (17, 18) are mainly the superior longitudinal fasciculus tistical comparison of the estimated MI in the sources’ activity. The and the arcuate fasciculus that connects the superior temporal results indicated a dense cortical network underpinning multi- gyrus with the dorsal prefrontal cortex (19, 20). This pathway has sensory integration, whereas a relatively sparser network sup- been associated with attentional shifting (21) and behavioral ports the identification of abstract audiovisual incongruences. choices for visual stimuli in both target-present and target-absent Musicians show greater connectivity than nonmusicians be- conditions, predicting choice accuracy (22). The cortical organi- tween distributed cortical areas, including a greater contribution zation of this large-scale network indicates that nonmusicians rely of the right temporal cortex for multisensory integration and the heavily on the processing of the corresponding visual cues for the left inferior frontal cortex for identifying abstract audiovisual in- integration of congruent audiovisual information. congruences. Moreover, the network organization of musicians In contrast, the large-scale network used by musicians for iden- indicates enhanced processing efficiency. tifying congruent audiovisual information shows significantly greater The behavioral results are in line with previously reported connectivity between distributed cortical sources. Direct connec- ones (9), indicating that musicians’ prior experience in musical tions between occipital and frontal sources contribute to multisen- reading is associated with a behavioral benefit. Previous psy- sory perception; nevertheless, the prominent role of right superior chophysical studies have reported similar results among non- and middle temporal gyri indicates an important role of auditory musicians, indicating that the identification of cross-modal information processing in this network. This result is in line with correspondences between pitch height and the height of a visual recent neurophysiological evidence, indicating that musicians show stimulus may be innate (13). In contrast, a recent study by our greater activation of the right middle and superior temporal gyri, group investigating the audiotactile integration (14) indicated which is correlated with increased sensitivity to acoustic features that similar rules binding auditory and tactile stimuli were not and enhanced selective attention to temporal features of speech intrinsically incorporated by the nonmusicians, but that musical (23). The comparison of the networks used by musicians and non- expertise was essential for reaching correct identification of musicians reveals significant differences in the connectivity of the audiotactile incongruences. temporal sources with precentral and frontal ones. This result seems The statistical comparison of the individual cortical connectivity to depict the audiovisual-motoric binding that musicians have highly matrices indicated that when nonmusicians were confronted with trained to read musical scores and to play their instrument (24). congruent audiovisual trials they activated a large-scale functional This binding was recently shown also by an MEG study of our group network that followed a mainly dorsal pathway. Although this (14), revealing how musical training molds the neural correlates of network consists of distributed sources, there is a prominent con- an integrative audiotactile response. Moreover, musicians show tribution of connections between occipitoparietal and frontal increased connectivity of the temporal sources with the prefrontal

Table 1. Global efficiency and density values of the significant networks of musicians and nonmusicians for each condition Condition Network organization parameter Musicians Nonmusicians

Congruent > incongruent Global efficiency 0.00063 0.00041 Density 0.00019 0.00016 Incongruent > congruent Global efficiency 0.00049 0.00026 Density 0.00022 0.00003

Paraskevopoulos et al. PNAS Early Edition | 3of6 Downloaded by guest on September 24, 2021 cortex, a result that is in line with diffusion tensor imaging (DTI) and ventral pathway for the long-range connections and relies studies, indicating increased functional anisotropy of the right ar- greatly on the contribution of the left inferior frontal cortex. The cuate fasciculus in musicians (25). In these studies, the increase of organization of this network indicates that musicians make functional anisotropy was attributed to increased myelination, greater use of the temporal sources (and auditory information caused by neural activity in fiber tracts during training. In ad- accordingly). The present results closely relate to previous results dition, the network organization parameters indicate that the reported by our group (9, 10). This analysis further revealed that functional connectivity pattern that musicians use to process musicians showed significantly greater activity than nonmusicians congruent audiovisual information shows greater efficiency and in superior frontal, superior temporal, and lingual gyrus when is denser than the corresponding network of the nonmusicians, confronted with incongruences. More importantly, the signifi- reflecting enhanced multisensory processing. cant role of inferior frontal cortex for response inhibition (31) The statistical analysis of the cortical connectivity of non- and for the identification of auditory pattern violations (32, 33) musicians underpinning abstract audiovisual identification (Fig. 3) has been repeatedly shown using several neuroimaging modali- revealed a sparser network, as indicatedalsobyitslowdensityand ties (34) not only for linguistic (35) but also for musical stimuli efficiency. This network connects mainly parietal and frontal areas, (36). In addition, musicians show enhanced gray matter density receiving input also from occipital (cuneus) and temporal (trans- (37) and enhanced connectivity of this region (38). Functional verse temporal gyrus) sources, indicating combined top-down and MRI evidence links this area to the processing of audiovisual bottom-up processing. The main connections of inferior parietal speech (39) as well as the enhanced visuospatial processing of and frontal sources closely correspond to the pathway subserving musicians (40) and thus, it may ideally serve the superior iden- visual spatial awareness. Its structural substrate is linked to superior tification of abstract audiovisual incongruences that musicians occipitofrontal fasciculus (26). This system seems to be dissociated show in this study. The network organization parameters reveal from auditory spatial processing (27). Nevertheless, our results that the network of musicians has enhanced density and pro- indicate significant functional connectivity with the transverse cessing efficiency. It has to be noted here that functional con- temporal gyrus, most likely supporting a bottom-up process un- nectivity as measured in the present study has been shown to derpinning a multisensory component (28). In addition, coac- closely relate to structural connectivity (41), nevertheless, func- tivation, and thus connectivity, in these areas has been correlated tional connectivity analysis also provides different and comple- with working memory tasks (29). These cognitive functions serve mentary information (42). an important role in identifying the abstract audiovisual incon- gruences presented in the paradigm of this study and may be Conclusion interpreted, in resemblance to the multisensory integration net- The present study revealed for the first time to our knowledge work, as indications that nonmusicians recruit stronger contri- the large-scale functional network underpinning audiovisual in- bution from the processing of visual information for identifying tegration as well as the effect of musical expertise on reorganizing the incongruences. this network. A novel approach for graph theoretical analysis of The group of musicians employs a large-scale cortical network MEG data revealed that nonmusicians rely heavily on the pro- to identify abstract audiovisual incongruences. This network cessing of corresponding visual clues for the integration of au- consists mainly of long-range occipitofrontal and parietofrontal diovisual information, whereas in contrast, musicians use more connections, having edges also to the right middle and superior temporal sources along with the left inferior frontal gyrus. Hence, temporal gyrus. Its general structure indicates the use of superior the results suggest that large-scale functional connectivity subserving longitudinal and arcuate fasciculus (30), resembling the network multisensory integration is reorganized due to expertise in a cogni- used by nonmusicians for the identification of congruent multi- tive domain that largely involves audiovisual integration, such as sensory information, although being more strongly lateralized to music reading, indicating that the corresponding cortical network the right hemisphere. The statistical analysis of the group dif- forms a dynamic system that can be affected by long-term training. ferences for this condition indicates the significantly greater contribution of the left inferior frontal cortex in the network Methods connectivity in musicians compared with nonmusicians. Specifi- Subjects. A new dataset was created for the present study from MEG mea- cally, this analysis indicates that musicians show enhanced con- surements that have been presented in two previous publications (9, 10). In the nectivity in a large-scale network that employs both the dorsal new dataset, we included the same number of musicians and nonmusicians for

Fig. 3. Cortical network underpinning audiovisual incongruency identification. (Upper) Statistical parametric maps of the significant networks for the in- congruent > congruent comparison. Networks presented are significant at P < 0.05, FDR corrected. The color scale indicates t values. (Lower) Node strength of the significant networks for each comparison. Strength is represented by node size.

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1510662112 Paraskevopoulos et al. Downloaded by guest on September 24, 2021 whom MRI data were available. Hence, 26 individuals participated in the MEG Data Analysis. present study, 13 musicians and 13 nonmusicians. Musicians (mean age = Source activity estimation. The Brain Electrical Source Analysis software (BESA 25.32; SD = 2.68; five males) were retrieved from the pool of students of the Research, version 6, Megis Software) was used for the preprocessing of the Music Conservatory of the University of Münster (mean musical training = MEG data. Artifacts due to blinks or movements were corrected by 17.23; SD = 3.81). Nonmusicians (mean age = 26.46; SD = 2.66; six males) had applying an adaptive artifact correction (44). The recorded data were sep- not received any formal musical education apart from compulsory school les- arated in epochs of 1 s, including a prestimulus interval of 200 ms. Epochs sons. All subjects were right handed according to the Edinburgh Handedness were baseline corrected using the interval from −100 to 0 ms. From each Inventory (43) and had normal hearing as evaluated by clinical audiometry. congruent trial only one randomly selected tone–number pair was included Subjects provided written informed consent before their participation in the in the analysis, thus producing an equal number of congruent and in- study. The study protocol was approved by the ethics committee of the congruent epochs. Data were filtered offline with a high-pass forward filter medical faculty of the University of Münster and the study was conducted of 1 Hz, a low-pass zero-phase filter of 30 Hz, and an additional notch filter according to the Declaration of Helsinki. at 50 Hz. Epochs containing signals larger than 2.5 picotesla in the MEG were considered as artifact contaminated and excluded from averaging. The Stimuli. Audiovisual congruent and incongruent stimuli were prepared by congruent and incongruent trials of each run were separately averaged. combining melodies that consisted of five tones with five images representing BESA MRI was used to construct individual FEM models. The individual the pitch height of each tone. Each image presented five white horizontal lines positions of the MEG sensors were coregistered via anatomical landmarks against a black background, similar to the staff lines in music notation. A blue (nasion, left and right preauricular points) to each subject’s structural MRI. disk (RGB color codes: red, 86; green, 126; and blue, 214) was then placed in Transformation to anterior commissure - posterior commissure and Talairach the middle of the screen on the horizontal axis and in one of the four spaces space was performed by direct 3D-spline interpolation of the original MRI. A between the lines. The five-tone melodies were constructed by the combination finite element model, based on the segmentation of four different head of four sinusoidal tones (F5, 698.46 Hz; A5, 880.46 Hz; C6, 1046.50 Hz; and tissues [i.e., scalp, skull, cerebrospinal fluid (CSF), and brain], was computed E6, 1318.51 Hz) with duration of 400 ms including 10-ms rise and decay time for each participant on the basis of their individual structural MRI and used (48,000 Hz, 16 bit). The interstimulus interval (stimulus offset to stimulus onset, as a volume conductor model. The conductivity value for the skin com- ISI) between each tone was set to 500 ms and the total duration of each melody partment was set to 0.33 S/m, for the skull compartment to 0.0042 S/m, for was 4 s. Eight different trials of five-tone melodies and five images were the brain to 0.33 S/m, and for the CSF to 0.79 S/m as proposed in ref. 45. The prepared for each condition (i.e., congruent and incongruent, respectively). use of realistic FEMs generated on the basis of the individual anatomy of For the audiovisual “congruent” trials, the melodies were combined with the each subject improves the source localization reliability of the MEG signals, corresponding image according to the rule “the higher the pitch, the higher the especially when the compartment of CSF is included in the model (46). position of the disk,” thus following a similar rule of ordinal relationship that Current density reconstructions (CDRs) were calculated on the neural re- the Western music notation system uses. During the acoustic presentation of sponses of each subject separately for each run’s congruent and incongruent

each tone, the corresponding disk was presented visually at the middle of the averages using the sLORETA method (12) as provided by BESA. This method is NEUROSCIENCE screen, at the exact same time, and for the same duration as the tone. During an unweighted minimum norm that is standardized by the resolution matrix. the ISI, a fixation cross was presented. The audiovisual “incongruent trials” Hence, it has the advantage of not needing an a priori definition of the number were prepared along the same principles as the congruent ones, except that of activated sources. Each individual’sCDRswerecalculatedforthecomplete one of the tones of the melodies was paired with an image that did not cor- response time window (i.e., 0–800 ms) of each condition, each run, and for each respond to the tone according to the above-mentioned rule. The incongruency sample point. The CDRs were exported from BESA as four-dimensional (4D) nifti was never in the first or last tone–number pair, but occurred at all of the other images and the images of each run were then averaged, resulting in one 4D three positions with equal probabilities (Fig. 1). image of 480 samples for each subject and each condition. The images were then processed with a mask that included only the gray matter and excluded Experimental Design. the brainstem and cerebellum to limit the source space. This procedure resulted MEG recordings. Evoked magnetic fields were recorded with a 275-channel in a source space of 863 voxels. whole-head system (OMEGA, CTF Systems) in a magnetically shielded room. Graph analysis. The CDR voxel time series were extracted in order for the con- Data were acquired continuously with a sampling rate of 600 Hz. Subjects nectivity matrices to be calculated, whereas a node of the network was were seated upright, and their head position was comfortably stabilized appointed to each voxel of the masked 4D images (resulting in 863 nodes). The with pads inside the MEG dewar. The auditory stimulation was delivered via Matlab R2010a (MathWorks) toolbox HERMES (47) was used for calculating the 60-cm-long silicon tubes at 60 dB SL above the individual hearing threshold 863 × 863 adjacency matrix from the voxel time series (i.e., the 4D images) of that was determined with an accuracy of at least 5 dB at the beginning of each subject and each condition was based on the algorithm of MI. MI quan- each MEG session for each ear. Visual stimuli were projected with an tifies the amount of information that is obtained for a random variable by Optoma EP783S DLP projector and a refresh rate of 60 Hz onto the back of a observing another. The main advantage of MI is that, being based on nonlinear semitransparent screen positioned ∼90 cm in front of the subjects’ nasion. probability distributions, it detects higher order correlations. Therefore, its re- The viewing angle ranged from −3.86° to 3.86° in the horizontal direction sult is not dependent on any specific model of the data (47). and from −1.15° to 1.15° in the vertical direction. The toolbox Network Based Statistic (NBS) (48) was used to statistically The congruent and incongruent trials were randomly interleaved in one identify significant connections in the graphs, using a general linear model run. The continuous recording was time locked to the presentation of the approach that compares the connectivity values of each node and for each second, third, and fourth tones/images of the audiovisual congruent patterns subject and condition. Specifically, a mixed model ANOVA with a between- and the presentation of the incongruent tone/image of the audiovisual in- subject factor group and a within-subject factor condition was designed to congruent patterns. This procedure resulted in an incongruent (deviant) to explore the main effect of condition within each group and the group × congruent (standard) ratio of 33.3%. Within 2.5 s after each trial, subjects condition interaction via planned contrasts. The significance level was set to had to answer via button pressings if the trial was congruent or incongruent P < 0.05 (or P < 0.001 when noted accordingly) corrected for multiple (right hand). During this intertrial interval, an image was presented to the comparisons via FDR correction. The visualization of the significant networks subjects reminding them which button represented each answer. Subjects as weighted graphs was performed using BrainNet Viewer (49). were exposed to four measurement runs, lasting ∼14.5 min each, with short To investigate the functional connectivity parameters and the efficacy of the breaks in between. The total number of trials for each condition was 104. networks, the graph properties of global efficiency, network density, and node MRI protocol. Before the experiment, T1-weighted MR images from each strength were calculated for the significant networks, as resulted from the NBS individual were obtained in a 3-Tesla scanner (Gyroscan Intera T30, Philips). analysis. The global efficiency is the average inverse shortest path length in the These anatomical scans served for constructing the individualized finite el- network, reflecting the level of global integration in the network. In other words, ement models (FEMs) that were then used in the source reconstruction. A global efficiency depicts the average number of nodes through which the in- total of 400 contiguous T1-weighted slices of 0.5-mm thickness in the sagittal formation flows to reach a distant point of the network. The fewer nodes that plane (TR = 7.33.64 ms, TE = 3.31 ms) were collected by a Turbo Field Echo intervene in the information flow, the greater the efficiency of the network. The acquisition protocol. The field of view was set to 300 × 300 mm with an in- node density is the fraction of present connections to possible connections, in- plane matrix of 512 × 512 setting defining the native voxel size at 0.5 × 0.58 × dicating how many of the available cortical areas actually contribute to the net- 0.58 mm3. Using SPM8 (Statistical Parametric Mapping, www.fil.ion.ucl.ac.uk) work, whereas the node strength sums the weights (i.e., t values in the present the intensity bias of the images was then regularized to account for intensity analysis) of links connected to the node, thus identifying the nodes that differences within each tissue, and the images were resliced to isotropic voxels show greater connectivity differences. These properties were calculated of 1.17 × 1.17 × 1.17 mm. using the Brain Connectivity Toolbox (50).

Paraskevopoulos et al. PNAS Early Edition | 5of6 Downloaded by guest on September 24, 2021 ACKNOWLEDGMENTS. This work was partly supported by the Deutsche research fellowship from the Research Committee of the Aristotle Univer- Forschungsgemeinschaft (Grant PA392/12-2) and partly by an internal sity of Thessaloniki (Award 31663/2014).

1. Ghazanfar AA, Schroeder CE (2006) Is neocortex essentially multisensory? Trends 28. Saint-Amour D, De Sanctis P, Molholm S, Ritter W, Foxe JJ (2007) Seeing voices: High- Cogn Sci 10(6):278–285. density electrical mapping and source-analysis of the multisensory mismatch nega- 2. Schroeder CE, Foxe J (2005) Multisensory contributions to low-level, ‘unisensory’ tivity evoked during the McGurk illusion. Neuropsychologia 45(3):587–597. processing. Curr Opin Neurobiol 15(4):454–458. 29. Friedman HR, Goldman-Rakic PS (1994) Coactivation of prefrontal cortex and inferior 3. Zamora-López G, Zhou C, Kurths J (2010) Cortical hubs form a module for multisen- parietal cortex in working memory tasks revealed by 2DG functional mapping in the sory integration on top of the hierarchy of cortical networks. Front Neuroinform 4:1. rhesus monkey. J Neurosci 14(5 Pt 1):2775–2788. 4. Greicius MD, Krasnow B, Reiss AL, Menon V (2003) Functional connectivity in the 30. Martino J, et al. (2013) Analysis of the subcomponents and cortical terminations of resting brain: A network analysis of the default mode hypothesis. Proc Natl Acad Sci the perisylvian superior longitudinal fasciculus: A fiber dissection and DTI tracto- – USA 100(1):253 258. graphy study. Brain Struct Funct 218(1):105–121. 5. Bullmore E, Sporns O (2009) Complex brain networks: Graph theoretical analysis of 31. Hampshire A, Chamberlain SR, Monti MM, Duncan J, Owen AM (2010) The role of the structural and functional systems. Nat Rev Neurosci 10(3):186–198. right inferior frontal gyrus: Inhibition and attentional control. Neuroimage 50(3): 6. Stam CJ, Reijneveld JC (2007) Graph theoretical analysis of complex networks in the 1313–1319. brain. Nonlinear Biomed Phys 1(1):3. 32. Maess B, Koelsch S, Gunter TC, Friederici AD (2001) Musical syntax is processed in 7. Supekar K, Musen M, Menon V (2009) Development of large-scale functional brain Broca’s area: A MEG study. Nat Neurosci 4(5):540–545. networks in children. PLoS Biol 7(7):e1000157. 33. Koelsch S (2006) Significance of Broca’s area and ventral premotor cortex for music- 8. Herholz SC, Zatorre RJ (2012) Musical training as a framework for brain plasticity: syntactic processing. Cortex 42(4):518–520. Behavior, function, and structure. Neuron 76(3):486–502. 34. Opitz B, Rinne T, Mecklinger A, von Cramon DY, Schröger E (2002) Differential con- 9. Paraskevopoulos E, Kuchenbuch A, Herholz SC, Pantev C (2012) Musical expertise induces audiovisual integration of abstract congruency rules. J Neurosci 32(50): tribution of frontal and temporal cortices to auditory change detection: fMRI and ERP – 18196–18203. results. Neuroimage 15(1):167 174. ’ 10. Paraskevopoulos E, Kuchenbuch A, Herholz SC, Pantev C (2014) Multisensory in- 35. Musso M, et al. (2003) Broca s area and the language instinct. Nat Neurosci 6(7): – tegration during short-term music reading training enhances both uni- and multi- 774 781. sensory cortical processing. J Cogn Neurosci 26(10):2224–2238. 36. Koelsch S, Gunter TC, Wittfoth M, Sammler D (2005) Interaction between syntax pro- 11. Lee H, Noppeney U (2011) Long-term music training tunes how the brain temporally cessing in language and in music: An ERP Study. J Cogn Neurosci 17(10):1565–1577. binds signals from multiple senses. Proc Natl Acad Sci USA 108(51):E1441–E1450. 37. Sluming V, et al. (2002) Voxel-based morphometry reveals increased gray matter 12. Pascual-Marqui RD (2002) Standardized low-resolution brain electromagnetic tomog- density in Broca’s area in male symphony orchestra musicians. Neuroimage 17(3): raphy (sLORETA): Technical details. Methods Find Exp Clin Pharmacol 24(Suppl D):5–12. 1613–1622. 13. Spence C, Deroy O (2013) How automatic are crossmodal correspondences? Conscious 38. Oechslin MS, Imfeld A, Loenneker T, Meyer M, Jäncke L (2009) The plasticity of the Cogn 22(1):245–260. superior longitudinal fasciculus as a function of musical expertise: A diffusion tensor 14. Kuchenbuch A, Paraskevopoulos E, Herholz SC, Pantev C (2014) Audio-tactile in- imaging study. Front Hum Neurosci 3:76. tegration and the influence of musical training. PLoS One 9(1):e85743. 39. Ojanen V, et al. (2005) Processing of audiovisual speech in Broca’s area. Neuroimage 15. Ungerleider LG, Mishkin M (1982) Two cortical visual systems. Analysis of Visual Behavior, 25(2):333–338. eds Ingle MA, Goodale MI, Masfield RJW (MIT Press, Cambridge, MA), pp 549–586. 40. Sluming V, Brooks J, Howard M, Downes JJ, Roberts N (2007) Broca’s area supports 16. Logothetis NK, Sheinberg DL (1996) Visual object recognition. Annu Rev Neurosci 19: enhanced visuospatial cognition in orchestral musicians. J Neurosci 27(14):3799–3806. – 577 621. 41. Greicius MD, Supekar K, Menon V, Dougherty RF (2009) Resting-state functional “ ” 17. Rauschecker JP, Tian B (2000) Mechanisms and streams for processing of what and connectivity reflects structural connectivity in the default mode network. Cereb “ ” – where in auditory cortex. Proc Natl Acad Sci USA 97(22):11800 11806. Cortex 19(1):72–78. 18. Rauschecker JP, Scott SK (2009) Maps and streams in the auditory cortex: Nonhuman 42. Damoiseaux JS, Greicius MD (2009) Greater than the sum of its parts: A review of primates illuminate human speech processing. Nat Neurosci 12(6):718–724. studies combining structural connectivity and resting-state functional connectivity. 19. Saur D, et al. (2008) Ventral and dorsal pathways for language. Proc Natl Acad Sci USA Brain Struct Funct 213(6):525–533. 105(46):18035–18040. 43. Oldfield R (1971) The assessment and analysis of handedness: the Edinburgh in- 20. Glasser MF, Rilling JK (2008) DTI tractography of the ’s language path- ventory. Neuropsychologia 9(1):97–113. ways. Cereb Cortex 18(11):2471–2482. 44. Ille N, Berg P, Scherg M (2002) Artifact correction of the ongoing EEG using spatial filters 21. Pammer K, Hansen P, Holliday I, Cornelissen P (2006) Attentional shifting and the role of based on artifact and brain signal topographies. JClinNeurophysiol19(2):113–124. the dorsal pathway in visual word recognition. Neuropsychologia 44(14):2926–2936. 45. Geddes LA, Baker LE (1967) The specific resistance of biological material—a com- 22. Donner TH, et al. (2007) Population activity in the human dorsal pathway predicts the accuracy of visual motion detection. J Neurophysiol 98(1):345–359. pendium of data for the biomedical engineer and physiologist. Med Biol Eng 5(3): – 23. Jantzen MG, Howe BM, Jantzen KJ (2014) Neurophysiological evidence that musical 271 293. training influences the recruitment of right hemispheric homologues for speech 46. Cho J-H, Vorwerk J, Wolters CH, Knösche TR (2015) Influence of the head model on – perception. Front Psychol 5:171. EEG and MEG source connectivity analyses. Neuroimage 110:60 77. 24. Schön D, Anton JLJL, Roth M, Besson M (2002) An fMRI study of music sight-reading. 47. Niso G, et al. (2013) HERMES: Towards an integrated toolbox to characterize func- Neuroreport 13(17):2285–2289. tional and effective brain connectivity. 11(4):405–434. 25. Wan CY, Schlaug G (2010) Music making as a tool for promoting brain plasticity across 48. Zalesky A, Fornito A, Bullmore ET (2010) Network-based statistic: Identifying differ- the life span. 16(5):566–577. ences in brain networks. Neuroimage 53(4):1197–1207. 26. Thiebaut de Schotten M, et al. (2005) Direct evidence for a parietal-frontal pathway 49. Xia M, Wang J, He Y (2013) BrainNet Viewer: A network visualization tool for human subserving spatial awareness in humans. Science 309(5744):2226–2228. brain connectomics. PLoS One 8(7):e68910. 27. Kong L, et al. (2014) Auditory spatial attention representations in the human cerebral 50. Rubinov M, Sporns O (2010) Complex network measures of brain connectivity: Uses cortex. Cereb Cortex 24(3):773–784. and interpretations. Neuroimage 52(3):1059–1069.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1510662112 Paraskevopoulos et al. Downloaded by guest on September 24, 2021